Custom HaarCascade fails

Hi guys,

I’m having trouble loading custom HaarCascade files.
If I try to load the default openMV frontalface cascade file, I get a ‘core dump: misaligned load’. I also tried this with a custom build haarcascade file, the bug persists.

My code:
image.HaarCascade('/flash/frontalface.cascade') #I also tried loading from '/sd', with the same result.

I tested this code on a Sipeed M1w Dock and a m5stickv with different Maixpy versions.

Has anyone got a fix???

For a complete description and error dump look at my github bug report.

Hi,
in MaixPy we doesn’t suggest use old haarcascade method, we suggest use this:
https://blog.sipeed.com/p/1338.html

Thank you for your reply Zepan.

I’m familiar with training tensor networks. however, I want to use the K210 to recognize number plates and I only have a database suited for training cascade files (https://github.com/openalpr/train-detector). In case of number plate detection cascades have proven very effective. Also, I want to use the KPU for OCR purposes.

I’ve put quite a lot of work in creating the right cascade file, so it’s a bummer to run into this bug. Number plate detection on a K210 would be quite awesome.